FiberNet

What is FiberNet? / 是什么

A research-grade Python toolkit for computational design of fiber network metamaterials. Complete closed-loop workflow: parametric generation, GPU-accelerated simulation, feature extraction, machine learning prediction, and reinforcement learning optimization — all through a unified Python API.

面向纤维网络超材料计算设计的 Python 工具包。从参数化生成、GPU 加速模拟、特征提取到机器学习预测与强化学习优化，完整闭环工作流通过统一 API 调用。

BMG-FDU, Fudan University / 复旦大学 生物大分子课题组

Why FiberNet? / 为什么选择

• One-line APIs — generate, simulate, predict, optimize in a single call
  一行代码完成生成、模拟、预测、优化
• GPU acceleration — large-scale structural simulation with millisecond latency
  GPU 加速大规模结构模拟
• Parametric design — continuous parameter space for inverse design and optimization
  连续参数空间支持逆向设计与优化
• Rich features — structural, topological, and mechanical descriptors out of the box
  内置结构、拓扑与力学特征描述符

One-Line API / 一行代码

import fibernet as fn

g = fn.pattern_2d(unit="honeycomb", grid=(4,4))
fn.show(g)
r = fn.simulate(g, mode="stretch", strain=1.5)
result = fn.predict_from_csv("data.csv", target="max_force")
best = fn.run_bayesian_optimization(obj, space, n_iter=50)

FEM in 3 Lines / 三行FEM

from fibernet.ml import BeamFrameFEM

solver = BeamFrameFEM(E=1e9, nu=0.3)
g = fn.pattern_2d(unit="honeycomb", box=(10, 10), grid=(4, 4), radius=0.05)
result = solver.stretch_test(g, target_stretch=2.0)

print(f"Max stress: {result['sigma_total'].max()/1e6:.1f} MPa")
print(f"Max displacement: {result['max_displacement']:.4f} m")

Euler-Bernoulli beam frame FEM with welded joints. Stretch and compression tests on any topology. / 欧拉-伯努利梁框架FEM，支持任意拓扑的拉伸与压缩测试。

Installation / 安装

pip install fibernet            # Core / 核心
pip install fibernet[full]      # ML + RL + Viz + Accel
pip install fibernet[ml]        # scikit-learn, pandas
pip install fibernet[rl]        # gymnasium, scikit-optimize
pip install fibernet[accel]     # taichi (GPU)
pip install fibernet[viz]       # pyvista (3D)

Platforms / 平台

Machine Learning / 机器学习

Built-in feature extraction produces 94-dimensional structural descriptors. predict_from_csv() handles the full pipeline: train/test split, nested cross-validation, model comparison, and visualization — all in one call.

内置特征提取生成94维结构描述符，一行完成训练、交叉验证、模型对比与可视化。

Reinforcement Learning / 强化学习

Built-in optimization engines over a continuous parametric action space of internal node displacements. Automatically closes the loop: generate new structures, simulate their properties, evaluate the objective, and update parameters.

内置优化引擎在连续参数空间中自动搜索最优结构，闭环完成生成、模拟、评估、更新。